AI Chats Can Be Used
U.S. lawyers are warning that interactions with AI chatbots may not be legally confidential after a recent court ruling that allowed such communications to be subject to discovery. (reuters.com) The guidance is being applied beyond consumer caution—law firms say employees treating chatbots as confidants risk creating discoverable records in litigation. (reuters.com)
A federal judge in New York has ruled that some chatbot conversations are not protected by legal privilege and can be turned over in court. (adamsandreese.com) The ruling came in *United States v. Heppner*, where prosecutors sought access to Bradley Heppner’s exchanges with Anthropic’s Claude after he used the chatbot while facing a criminal investigation. Judge Jed Rakoff of the Southern District of New York held that the chats were not covered by attorney-client privilege or the work-product doctrine. (adamsandreese.com) Law firms began warning clients on April 15 that the decision could reach beyond one defendant and one case. Reuters reported that New York firm Sher Tremonte added contract language saying that sharing a lawyer’s advice or communications with a chatbot could waive privilege. (money.usnews.com) Attorney-client privilege is the rule that usually shields private communications between lawyers and clients made for legal advice. The work-product doctrine separately protects material prepared for litigation, but Rakoff said a public chatbot did not qualify for either protection on the facts before him. (crowell.com) The court’s reasoning turned on a simple point: Heppner was talking to a third-party software service, not directly to his lawyers. Several law firm analyses said the opinion treated the chatbot like any other outside recipient that can break confidentiality when a user voluntarily shares sensitive material. (foxrothschild.com) (goodwinlaw.com) That warning now reaches ordinary workplace use, not just criminal defense strategy. Bloomberg Law said the logic could apply to enterprise tools that let employees ask legal or compliance questions without a lawyer directly supervising the exchange. (news.bloomberglaw.com) The privacy issue is separate from the courtroom issue, but it points in the same direction. OpenAI says content from consumer services such as ChatGPT may be used to improve models unless users opt out, while business offerings are not used for training by default unless customers explicitly opt in. (openai.com 1) (openai.com 2) That means a chat can be stored by a provider, discoverable by an opponent, or both, depending on the product and the lawsuit. Lawyers quoted by Reuters said employees who treat chatbots like private confidants may be creating records that later have to be produced in civil litigation or criminal cases. (insurancejournal.com) The immediate lesson from the Heppner ruling is narrower than “all AI chats are public,” but broader than many users assumed. If a person types legal strategy, internal facts, or attorney advice into a public chatbot, a court may treat that text like any other disclosure to an outside party. (adamsandreese.com)
